def test_padded_align_odd() -> None:
"""
Test the padded cross-correlation method with two gaussian pulses of odd length.
"""
# make some time arrays and define a time delta
t = np.linspace(-2 * np.pi, 2 * np.pi, 100)
dt = 3.675 * (t[1] - t[0])
t = np.linspace(-1, 1, 199, endpoint=False)
# construct two identical gaussian pulses
x = np.real(signal.gausspulse(t, fc=5))
y = np.real(signal.gausspulse(t + dt, fc=5))
# compute the delay without any upsampling
d = align.xcorr_delay(x, y, 30)
# and apply the delay
y = align.apply_delay(y, d)
# and test the top-level call
ynew = align.align(x, y, method="xcorr", factor=30)
# and check that the alignment is within a certain known accuracy
assert np.std(np.abs(x - y)) < 0.005
assert np.std(np.abs(y - ynew)) < 5e-5
def test_fft_align() -> None:
"""
Test FFT phase shift alignment
"""
for N in [178, 179, 200, 201]:
# make some time arrays and define a time delta
t = np.linspace(-2 * np.pi, 2 * np.pi, N // 2)
dt = 3.675 * (t[1] - t[0])
t = np.linspace(-1, 1, N, endpoint=False)
true = dt / (t[1] - t[0])
# construct two identical gaussian pulses
x = np.real(signal.gausspulse(t, fc=5))
y = np.real(signal.gausspulse(t + dt, fc=5))
# compute the delay with upsampling
delay = align.fft_delay(x, y)
# and apply the delay
y = align.apply_delay(y, delay)
# use the top-level call
ynew = align.align(x, y, method="fft")
# and check that the alignment is within a certain known accuracy
assert np.abs(true - delay) < 1e-3
assert np.mean(np.abs(x - y)) < 1e-3
assert np.mean(np.abs(x - ynew)) < 2e-4
assert np.mean(np.abs(y - ynew)) < 3e-5
def test_gauss_align() -> None:
"""
Test Gaussian interpolation of unpadded cross correlation.
"""
# make some time arrays and define a time delta
t = np.linspace(-2 * np.pi, 2 * np.pi, 100)
dt = 3.675 * (t[1] - t[0])
t = np.linspace(-1, 1, 200, endpoint=False)
# construct two identical gaussian pulses
x = np.real(signal.gausspulse(t, fc=5))
y = np.real(signal.gausspulse(t + dt, fc=5))
# compute the delay with upsampling
d_padded = align.xcorr_delay(x, y, 30)
# compute the delay without any upsampling
d = align.xcorr_delay(x, y, fit="gauss")
# compute the new signal
y = align.apply_delay(y, d)
# and test the top-level call
ynew = align.align(x, y, method="xcorr", factor=30)
# and check that the alignment is within a certain known accuracy
assert np.abs(d - d_padded) < 0.05
assert np.mean(np.abs(x - y)) < 7e-4
assert np.mean(np.abs(x - ynew)) < 7e-4
assert np.mean(np.abs(y - ynew)) < 5e-5
def test_correlate(self):
# create noise and a glitch template at 1000 Hz
noise = self.TEST_CLASS(numpy.random.normal(size=16384 * 64),
sample_rate=16384,
epoch=-32).zpk([], [1], 1)
glitchtime = -16.5
glitch = self.TEST_CLASS(signal.gausspulse(numpy.arange(
-1, 1, 1. / 16384),
bw=100),
sample_rate=16384,
epoch=glitchtime - 1)
# check that, without a signal present, we only see background
snr = noise.correlate(glitch, whiten=True)
tmax = snr.times[snr.argmax()]
assert snr.size == noise.size
assert not numpy.isclose(tmax.value, glitchtime)
nptest.assert_almost_equal(snr.mean().value, 0.0, decimal=1)
nptest.assert_almost_equal(snr.std().value, 1.0, decimal=1)
# inject and recover the glitch
data = noise.inject(glitch * 1e-4)
snr = data.correlate(glitch, whiten=True)
tmax = snr.times[snr.argmax()]
nptest.assert_almost_equal(tmax.value, glitchtime)
def test_gate(self):
# generate Gaussian noise with std = 0.5
noise = self.TEST_CLASS(numpy.random.normal(scale=0.5, size=16384*64),
sample_rate=16384, epoch=-32)
# generate a glitch with amplitude 20 at 1000 Hz
glitchtime = 0.0
glitch = signal.gausspulse(noise.times.value - glitchtime,
bw=100) * 20
data = noise + glitch
# check that the glitch is at glitchtime as expected
tmax = data.times.value[data.argmax()]
nptest.assert_almost_equal(tmax, glitchtime)
# gating method will be called with whiten = False to decouple
# whitening method from gating method
tzero = 1.0
tpad = 1.0
threshold = 10.0
gated = data.gate(tzero=tzero, tpad=tpad, threshold=threshold,
whiten=False)
# check that the maximum value is not within the region set to zero
tleft = glitchtime - tzero
tright = glitchtime + tzero
assert not tleft < gated.times.value[gated.argmax()] < tright
# check that there are no remaining values above the threshold
assert gated.max() < threshold
def test_gate(self):
# generate Gaussian noise with std = 0.5
noise = self.TEST_CLASS(numpy.random.normal(scale=0.5,
size=16384 * 64),
sample_rate=16384,
epoch=-32)
# generate a glitch with amplitude 20 at 1000 Hz
glitchtime = 0.0
glitch = signal.gausspulse(noise.times.value - glitchtime, bw=100) * 20
data = noise + glitch
# check that the glitch is at glitchtime as expected
tmax = data.times.value[data.argmax()]
nptest.assert_almost_equal(tmax, glitchtime)
# gating method will be called with whiten = False to decouple
# whitening method from gating method
tzero = 1.0
tpad = 1.0
threshold = 10.0
gated = data.gate(tzero=tzero,
tpad=tpad,
threshold=threshold,
whiten=False)
# check that the maximum value is not within the region set to zero
tleft = glitchtime - tzero
tright = glitchtime + tzero
assert not tleft < gated.times.value[gated.argmax()] < tright
# check that there are no remaining values above the threshold
assert gated.max() < threshold
def cwt_scipy():
st = obspy.read()[0].data
t = np.linspace(-1, 1, 200, endpoint=False)
sig = np.cos(2 * np.pi * 7 * t) + signal.gausspulse(t - 0.4, fc=2)
print(np.shape(st), st.dtype)
print(np.shape(sig), sig.dtype)
widths = np.arange(1, 100)
cwtmatr = abs(signal.cwt(st, signal.morlet2, widths))
# t = np.linspace(0, dt * npts, npts)
# x, y = np.meshgrid(t, np.logspace(np.log10(1), np.log10(50), cwtmatr.shape[0]))
# print(np.shape(y))
# plt.pcolormesh(x, y, np.abs(cwtmatr), cmap=obspy_sequential)
# plt.yscale('log')
plt.imshow(cwtmatr, aspect='auto')
# plt.imshow(cwtmatr, extent=[-1, 1, 1, 30], cmap='PRGn', aspect='auto',
# vmax=abs(cwtmatr).max(), vmin=-abs(cwtmatr).max())
plt.savefig('cwt_scipy.png')
plt.close()
def main():
t = np.linspace(-1, 1, 100)
def f(t):
return np.sin(
np.pi * t) + 0.1 * np.cos(7 * np.pi * t + 0.3) + 0.2 * np.cos(
24 * np.pi * t) + 0.3 * np.cos(12 * np.pi * t + 0.5)
fig, axes = plt.subplots(1, 3, figsize=(12, 3))
axes[0].plot(t, signal.gausspulse(t, fc=5, bw=0.5))
axes[0].set_title("gausspulse")
t *= 5 * np.pi
axes[1].plot(t, signal.sawtooth(t))
axes[1].set_title("chirp")
axes[2].plot(t, signal.square(t))
axes[2].set_title("gausspulse")
t = np.linspace(0, 4, 400)
plt.plot(t, f(t))
# 中值滤波函数
plt.plot(t,
signal.medfilt(f(t), kernel_size=55),
linewidth=2,
label="medfilt")
# 维纳滤波函数
plt.plot(t, signal.wiener(f(t), mysize=55), linewidth=2, label="wiener")
plt.legend()
plt.show()
def test_whiten(self):
# create noise with a glitch in it at 1000 Hz
noise = self.TEST_CLASS(numpy.random.normal(loc=1,
scale=.5,
size=16384 * 64),
sample_rate=16384,
epoch=-32).zpk([], [0], 1)
glitchtime = 0.5
glitch = signal.gausspulse(noise.times.value - glitchtime,
bw=100) * 1e-4
data = noise + glitch
# when the input is stationary Gaussian noise, the output should have
# zero mean and unit variance
whitened = noise.whiten(detrend='linear')
assert whitened.size == noise.size
nptest.assert_almost_equal(whitened.mean().value, 0.0, decimal=2)
nptest.assert_almost_equal(whitened.std().value, 1.0, decimal=2)
# when a loud signal is present, the max amplitude should be recovered
# at the time of that signal
tmax = data.times[data.argmax()]
assert not numpy.isclose(tmax.value, glitchtime)
whitened = data.whiten(detrend='linear')
tmax = whitened.times[whitened.argmax()]
nptest.assert_almost_equal(tmax.value, glitchtime)
def gen_samples(self, time, seq_shape, kwargs):
# TODO: MAKE IT DEPENDENT OF PARAMETERS, SAMPLE OUTSIDE
if seq_shape == "constant":
samples = np.ones(shape=time.shape) * kwargs["amp"]
params = np.array([
kwargs["amp"],
])
return samples, params
elif seq_shape in ["sinusoidal", "square", "sawtooth"]:
samples = 0
if seq_shape == "sinusoidal":
samples = np.sin(2 * np.pi * kwargs["freq"] * time +
kwargs["phase"]) * kwargs["amp"]
elif seq_shape == "square":
samples = ss.square(2 * np.pi * kwargs["freq"] * time +
kwargs["phase"]) * kwargs["amp"]
elif seq_shape == "sawtooth":
samples = ss.sawtooth(2 * np.pi * kwargs["freq"] * time +
kwargs["phase"]) * kwargs["amp"]
params = np.array([kwargs["amp"], kwargs["freq"], kwargs["phase"]])
return samples, params
elif seq_shape == "gaussian_pulses":
amp = np.random.uniform(low=self.amp_range[0],
high=self.amp_range[1])
freq = np.random.uniform(low=self.freq_range[0],
high=self.freq_range[1])
pulse_width = np.random.uniform(low=self.pulse_width_range[0],
high=self.pulse_width_range[1])
params = np.array([0])
samples = ss.gausspulse(time, freq, pulse_width)
def test_whiten(self):
# create noise with a glitch in it at 1000 Hz
noise = self.TEST_CLASS(
numpy.random.normal(loc=1, scale=.5, size=16384 * 64),
sample_rate=16384, epoch=-32).zpk([], [0], 1)
glitchtime = 0.5
glitch = signal.gausspulse(noise.times.value - glitchtime,
bw=100) * 1e-4
data = noise + glitch
# when the input is stationary Gaussian noise, the output should have
# zero mean and unit variance
whitened = noise.whiten(detrend='linear')
assert whitened.size == noise.size
nptest.assert_almost_equal(whitened.mean().value, 0.0, decimal=2)
nptest.assert_almost_equal(whitened.std().value, 1.0, decimal=2)
# when a loud signal is present, the max amplitude should be recovered
# at the time of that signal
tmax = data.times[data.argmax()]
assert not numpy.isclose(tmax.value, glitchtime)
whitened = data.whiten(detrend='linear')
tmax = whitened.times[whitened.argmax()]
nptest.assert_almost_equal(tmax.value, glitchtime)
def make_noise(text_length, filename):
# checking audio file to match with
nb_samples = audiofile.samples(filename)
sampling_rate = audiofile.sampling_rate(filename) # in Hz
# Generate random logarithmic noise
noise = np.random.lognormal(0, 1, nb_samples)
noise /= np.amax(np.abs(noise))
# Filter with bandpass filter
nyq = 0.5 * sampling_rate
low = text_length / nyq
high = (50 + text_length) / nyq
order = 1
b, a = signal.butter(order, [low, high], btype="band")
filtered_noise = signal.lfilter(b, a, noise)
# create Gaussian enveloppe
t = np.linspace(start=-0.5, stop=0.5, num=nb_samples, endpoint=False)
i, e = signal.gausspulse(t, fc=5, retenv=True, bwr=-6.0)
out_signal = np.multiply(filtered_noise, e)
out_signal *= 30
# write audio file
audiofile.write(filename + ".noise.wav", out_signal, sampling_rate)
print("text length was :", text_length)
def feedforward_modelbased(delay):
t = np.linspace(-1, 1, in_len, endpoint=False)
i, q, e = signal.gausspulse(t, fc=5, retquad=True, retenv=True)
temp = np.zeros((1, in_len))
temp[0, :] = i
out = np.zeros((1, in_len))
LEN = int(delay.shape[1] / 2)
for i in range(LEN):
scale = delay[0][LEN + i] - 0.5
# Add scale noise
scale += 1e-7 * np.random.randn()
# Add delay and scaling
delay_scaled = scale * np.roll(temp, int(
(in_len * delay).round()[0][i]))
# Add exponential damping
delay_scaled = np.multiply(
delay_scaled, 100.0 * delay[0][-1] /
np.linspace(1, in_len, in_len, endpoint=True))
# Add all delayed scaled pulses together
out += delay_scaled
# Adding noise
out += 1e-5 * np.random.randn(1, in_len)
return out
def test_save_loudest_tile_features():
# prepare input data
channel = GW.channels[0]
noise = TimeSeries(numpy.random.normal(loc=1, scale=.5, size=16384 * 68),
sample_rate=16384,
epoch=-34).zpk([], [0], 1)
glitch = TimeSeries(signal.gausspulse(numpy.arange(-1, 1, 1. / 16384),
bw=100),
sample_rate=16384,
epoch=-1) * 1e-4
in_ = noise.inject(glitch)
_, _, _, qgram, _, _, _ = core.scan(gps=0,
channel=channel,
xoft=in_,
resample=4096,
fftlength=8)
# test loudest tiles
channel.save_loudest_tile_features(qgram, correlate=glitch)
assert channel.Q == numpy.around(qgram.plane.q, 1)
assert channel.energy == numpy.around(qgram.peak['energy'], 1)
assert channel.snr == numpy.around(qgram.peak['snr'], 1)
assert channel.t == numpy.around(qgram.peak['time'], 3)
assert channel.f == numpy.around(qgram.peak['frequency'], 1)
assert channel.corr == numpy.around(glitch.max().value, 1)
assert channel.delay == 0.0
assert channel.stdev == glitch.std().value
def plotGpulse(self, amp, freq):
self.amp = amp
self.freq = freq
self.Fs: int = 44100
self.x = arange(0, 1, 1 / self.Fs)
self.y = (amp / 10) * gausspulse((self.x, freq))
self.fullPlot.plot(self.x, self.y, pen=self.penR)
self.zoomedPlot.plot(self.x, self.y, pen=self.penB)
def wavelet():
t = np.linspace(-1, 1, 200, endpoint=False)
sig = np.cos(2 * np.pi * 7 * t) + signal.gausspulse(t - 0.4, fc=2)
widths = np.arange(1, 31)
cwtmatr = signal.cwt(sig, signal.ricker, widths)
plt.imshow(cwtmatr, extent=[-1, 1, 1, 31], cmap='PRGn', aspect='auto', vmax=abs(cwtmatr).max(), vmin=-abs(cwtmatr).max())
plt.show()
return
def generate_SineGaussianBlip(fc, bw, duration):
#returns sine gaussian blip with bandwidth = bw, centre frequency = fc
t = np.linspace(-duration/2, duration/2, duration*4096)
sine_gaussian = signal.gausspulse(t, fc=fc, bw=bw)
waveform = TimeSeries(sine_gaussian, delta_t=1/4096)
return waveform
def test_gausspulse(self, time_data_gen, num_samps, fc):
cpu_time, gpu_time = time_data_gen(0, 10, num_samps)
cpu_pwm = signal.gausspulse(cpu_time, fc, retquad=True, retenv=True)
gpu_pwm = cp.asnumpy(
cusignal.gausspulse(gpu_time, fc, retquad=True, retenv=True))
assert array_equal(cpu_pwm, gpu_pwm)
def test_gausspulse(num_samps, fc):
cpu_time = np.linspace(0, 10, num_samps)
gpu_time = cp.asarray(cpu_time)
cpu_pwm = signal.gausspulse(cpu_time, fc, retquad=True, retenv=True)
gpu_pwm = cp.asnumpy(
cusignal.gausspulse(gpu_time, fc, retquad=True, retenv=True))
assert array_equal(cpu_pwm, gpu_pwm)
def set_gaussian_excitation (self, plot=False, bw=.5, bwr=-6, num_sigmas=3):
'''Define the ultrasound excitation signal.
`bw` is the fractional bandwidth (between 0 and 1) of the Gaussian
modulating signal relative to `tx.center_frequency.
`bwr` is the reference level at which the fractional bandwidth is
calculated (dB).
(See scipy.signal.gausspulse.)
`num_sigmas` is how many standard deviations outwards from the
center to sample in either direction.'''
from scipy.signal import gausspulse
import matplotlib.pyplot as plt
fc = self.tx.center_frequency
fs = self.sampling_frequency
# https://github.com/scipy/scipy/blob/14142ff70d84a6ce74044a27919c850e893648f7/scipy/signal/waveforms.py#L232
# exp(-a t^2) <-> sqrt(pi/a) exp(-pi^2/a * f^2) = g(f)
bwr = -6
ref = pow (10.0, bwr / 20.0)
a = -(np.pi * fc * bw) ** 2 / (4.0 * np.log (ref))
time_sigma = 1 / np.sqrt (2 * a)
time = np.arange (-num_sigmas * time_sigma,
num_sigmas * time_sigma, 1 / fs)
if plot:
# Plot the excitation signal
excitation, envelope = gausspulse(
time, fc=fc, bw=bw, bwr=bwr, retenv=True)
plt.plot(envelope, 'g--', excitation)
plt.title ('Excitation Signal', fontsize=20, pad=20)
plt.xlabel (r'Time [$\mu$s]', fontsize=14, labelpad=10)
plt.gca ().set_facecolor ('white')
xticks = plt.xticks ()[0]
plt.gca ().set_xticklabels ([f'{i*1e6/fs:.2f}' for i in xticks])
plt.grid (False, axis='y')
plt.grid (True, axis='x', linestyle=':', color='gray')
plt.yticks ([])
plt.show ()
else:
excitation = gausspulse(time, fc=fc, bw=bw, bwr=bwr)
# Set the excitation signal
self.excitation = self.new_tensor (excitation)
def create_gaussian_pulse(center_freq, samp_freq, bandwidth=2/3, bwr=-6,
tpr=-60, baseband=False, viz=False):
"""
Modification to `scipy.signal.gausspulse`:
https://github.com/scipy/scipy/blob/v0.14.0/scipy/signal/waveforms.py#L161
Their version doesn't let you set a center frequency of 0 (baseband case).
:return:
"""
# compute within [+/- cutoff]
t_cutoff = gausspulse('cutoff', fc=center_freq, bw=bandwidth, bwr=bwr,
tpr=tpr)
# gaussian coeff (Time domain compressive beam forming of ultrasound signals, David)
denom = 2 * np.pi * bandwidth * center_freq
tv = -(8 * np.log(10 ** (bwr / 20)) / (denom * denom))
alpha = 1. / 2 / tv
# compute pulse
t_vals = np.arange(-1. * t_cutoff, t_cutoff, 1 / samp_freq)
pulse = np.exp(-alpha * t_vals * t_vals)
if not baseband:
pulse *= np.cos(2 * np.pi * center_freq * t_vals)
if viz:
f, (ax1, ax2) = plt.subplots(2,1)
ax1.set_title("Gaussian pulse")
ax1.plot(t_vals, pulse)
ax1.grid()
ax1.autoscale(enable=True, axis='x', tight=True)
ax1.set_xlabel("Time [s]")
# ax1.get_xaxis().set_ticklabels([])
if baseband:
f_vals = np.linspace(-center_freq, center_freq, num=1000)
H = gauss_ft(f_vals, alpha)
ax2.plot(f_vals, 20*np.log10(np.abs(H)))
else:
f_vals = np.linspace(-2*center_freq, 2*center_freq, num=1000)
H = gauss_ft(f_vals, alpha, fc=center_freq)
ax2.plot(f_vals, 20*np.log10(np.abs(H)),
label="pulse")
ax2.axvline(x=center_freq, c='r', label="center frequency")
ax2.legend()
ax2.grid()
ax2.set_xlabel("Frequency [Hz]")
ax2.autoscale(enable=True, axis='x', tight=True)
ax2.set_ylabel("dB")
return pulse, t_vals, alpha
def test_poly_align() -> None:
"""
Test polynomial interpolation of unpadded cross correlation.
"""
# make some time arrays and define a time delta
t = np.linspace(-2 * np.pi, 2 * np.pi, 100)
dt = 3.675 * (t[1] - t[0])
t = np.linspace(-1, 1, 200, endpoint=False)
# construct two identical gaussian pulses
x = np.real(signal.gausspulse(t, fc=5))
y = np.real(signal.gausspulse(t + dt, fc=5))
# compute the delay with upsampling
d_padded = align.xcorr_delay(x, y, 30)
# compute the delay without any upsampling
d = align.xcorr_delay(x, y, fit="poly")
# and check that the alignment is within a certain known accuracy
assert np.abs(d - d_padded) < 0.05
def test_simple_align() -> None:
"""
Test the basic cross-correlation method with two gaussian pulses.
"""
# make some time arrays and define a time delta
t = np.linspace(-2 * np.pi, 2 * np.pi, 100)
dt = 3.675 * (t[1] - t[0])
t = np.linspace(-1, 1, 2 * 100, endpoint=False)
# construct two identical gaussian pulses
x = np.real(signal.gausspulse(t, fc=5))
y = np.real(signal.gausspulse(t + dt, fc=5))
# compute the delay without any upsampling
d = align.xcorr_delay(x, y)
# and apply the delay
y = align.apply_delay(y, d)
# and check that the alignment is within a certain accuracy
assert np.std(x - y) < 0.06
def wavelet():
t = np.linspace(-1, 1, 200, endpoint=False)
sig = np.cos(2 * np.pi * 7 * t) + signal.gausspulse(t - 0.4, fc=2)
widths = np.arange(1, 31)
cwtmatr = signal.cwt(sig, signal.ricker, widths)
plt.imshow(cwtmatr,
extent=[-1, 1, 1, 31],
cmap='PRGn',
aspect='auto',
vmax=abs(cwtmatr).max(),
vmin=-abs(cwtmatr).max())
plt.show()
return
def run(self, time, frq, bw=0.5, bwr=-6, ampl=1, retquad=False):
"""
Method to generate a sine signal for a given
time array, a frequency and amplitude.
"""
if retquad is False:
signal = gausspulse(time,
fc=frq,
bw=bw,
bwr=bwr,
retquad=False,
retenv=False)
elif retquad is True:
_, signal = gausspulse(time,
fc=frq,
bw=bw,
bwr=bwr,
retquad=True,
retenv=False)
signal = signal * ampl
signal_center = np.argmin(np.abs(time))
signal = signal.reshape(-1, 1)
return (signal, signal_center)
def run():
import seg1d
import numpy as np
import matplotlib.pylab as plt
import scipy.signal as signal
# create an array of data
x = np.linspace(-1, 1, 2000)
# get an array of data from a Gaussian pulse
targ = signal.gausspulse(x, fc=5)
# define a segment within the pulse to use as reference
t_s, t_e = 950, 1050
# cut a segment out to use as a reference data
refData = [{'gauss': targ[t_s:t_e]}]
targData = {'gauss': targ}
refWeights = {'gauss': 1}
### define some test parameters
minWin = 98 # minimum percent to scale down reference data
maxWin = 105 # maximum percent to scale up reference data
sizeStep = 1 # step to use for correlating reference to target data
# call the segmentation algorithm
segments = seg1d.segment_data(refData, targData, refWeights, minWin,
maxWin, sizeStep)
print(np.around(segments, decimals=7))
plt.figure(figsize=(15, 4))
# plot the full pulse
plt.plot(x, targ, linewidth=6, alpha=0.2, label='Target')
# plot the original reference segment
plt.plot(x[t_s:t_e],
targ[t_s:t_e],
linewidth=8,
alpha=0.5,
label='Reference')
# plot all segments found
seg_num = 1
for s, e, c in segments:
plt.plot(x[s:e],
targ[s:e],
dashes=[0.5, 0.5],
linewidth=4,
alpha=0.8,
label='Segment {}'.format(seg_num))
plt.legend()
plt.show()
def ping(self, positions, delays=None, show=False):
"""Send delayed pings at the given positions. Positions are relative to the center of
the field.
Args:
positions: Positions to send and receive pings, relative to the center of the field.
delays: Delay of the pings.
show: Animate the simulation.
Returns:
Returned signals.
"""
# Reset field to allow for repeated execution.
self.field.pressure.values = np.zeros_like(self.field.pressure.values)
self.field.velocity_x.values = np.zeros_like(self.field.velocity_x.values)
self.field.velocity_y.values = np.zeros_like(self.field.velocity_y.values)
self.field.step = 0
if delays is None:
delays = np.zeros(len(positions))
for position, delay in zip(positions, delays):
self.field.pressure.add_boundary(
self.field.get_point_region(
position=(self.field.x.vector[len(self.field.x.vector) // 2] + position,
max(self.field.y.vector))),
value=sg.gausspulse(self.field.t.vector - self.ping_pre_delay - delay,
self.ping_frequency, self.ping_bandwidth),
additive=True)
self.field.pressure.add_output(
self.field.get_point_region(
position=(self.field.x.vector[len(self.field.x.vector) // 2] + position,
max(self.field.y.vector))))
if show:
animator = fd.Animator2D(field=self.field, scale=0.1)
animator.show_boundaries = False
animator.show_materials = False
animator.show_output = False
animator.start_simulation()
else:
self.field.simulate()
if self.postprocessor:
return self.postprocessor(
[output.mean_signal for output in self.field.pressure.outputs])
else:
return [output.mean_signal for output in self.field.pressure.outputs]
def test_whiten(self):
# create noise with a glitch in it at 1000 Hz
noise = self.random.zpk([], [0], 1)
glitchtime = 0.5
glitch = signal.gausspulse(noise.times.value + glitchtime,
bw=100) * 1e-4
data = noise + glitch
# whiten and test that the max amplitude is recovered at the glitch
tmax = data.times[data.argmax()]
self.assertNotAlmostEqual(tmax.value, -glitchtime)
whitened = data.whiten(2, 1)
self.assertEqual(noise.size, whitened.size)
self.assertAlmostEqual(whitened.mean().value, 0.0, places=4)
tmax = whitened.times[whitened.argmax()]
self.assertAlmostEqual(tmax.value, -glitchtime)
def plotGpulse(self, amp, freq):
self.Fs: int = 44100
self.x = arange(0, 1, 1 / self.Fs)
self.y = (amp / 10) * gausspulse((self.x, freq))
self.plot = self.PlotView.plotItem
self.vBox = self.plot.vb
self.vBox.setLimits(xMin=-0, yMin=-2, xMax=1.3, yMax=2)
self.vBox.setBackgroundColor("w")
self.plot.addLegend()
self.plot.showGrid(x=True, y=True)
self.plot.plot(self.x,
self.y,
pen=mkPen('b', width=1),
name=str(amp) + "Gaussian Pulse(2π" + str(freq))
def test_whiten(self):
# create noise with a glitch in it at 1000 Hz
noise = self.random.zpk([], [0], 1)
glitchtime = 0.5
glitch = signal.gausspulse(
noise.times.value + glitchtime, bw=100) * 1e-4
data = noise + glitch
# whiten and test that the max amplitude is recovered at the glitch
tmax = data.times[data.argmax()]
self.assertNotAlmostEqual(tmax.value, -glitchtime)
whitened = data.whiten(2, 1)
self.assertEqual(noise.size, whitened.size)
self.assertAlmostEqual(whitened.mean(), 0.0, places=5)
tmax = whitened.times[whitened.argmax()]
self.assertAlmostEqual(tmax.value, -glitchtime)
def gausspulse(mean, std, start, end):
#t = np.linspace(-10, 10, 2 * 100, endpoint=False)
gausspulse = np.zeros((data_points))
num_data_points = int(data_points * (end - start))
start_zeros = np.zeros((int(data_points * start)))
t = np.linspace(-1, 1, num_data_points)
#t_year = np.linspace(-1, 1, num_minutes_in_year)
i, q, e = signal.gausspulse(t, fc=2, retquad=True, retenv=True)
i *= std
i += mean
final = np.append(start_zeros, i)
trailing_zeros = np.zeros(np.shape(gausspulse)[0] - np.shape(final)[0])
final2 = np.append(final, trailing_zeros)
gausspulse += final2
#plt.plot(t_year,gausspulse)
return gausspulse
def scalogram_example_1():
if True:
#N = 30
N = 1000
t = np.linspace(-1, 1, N, endpoint=False)
sig = np.cos(2 * np.pi * 7 * t) + signal.gausspulse(t - 0.4, fc=2)
else:
# A signal given.
sig = ...
N = len(sig)
t = np.linspace(0, 1 / Fs * N, N + 1)[:N]
widths = np.arange(1, N + 1)
#widths = np.arange(1, 500 + 1)
# Compute the CWT.
cwtmatr = signal.cwt(sig, signal.ricker,
widths) # cwtmatr.shape = (len(widths), len(sig)).
# Scalogram: a spectrogram for wavelets. (???)
plt.figure()
plt.pcolormesh(t, widths, 20 * np.log10(np.abs(cwtmatr)))
#plt.pcolormesh(t, widths[:500], 20 * np.log10(np.abs(cwtmatr[:500,:])))
plt.title('Scalogram')
#plt.xlabel('?')
#plt.ylabel('?')
plt.tight_layout()
# Plot the CWT.
plt.figure()
#plt.imshow(cwtmatr, extent=[t[0], t[-1], widths[0], widths[-1]], cmap='PRGn', aspect='auto', vmax=abs(cwtmatr).max(), vmin=-abs(cwtmatr).max())
plt.imshow(cwtmatr,
extent=[t[0], t[-1], widths[-1], widths[0]],
cmap='PRGn',
aspect='auto',
vmax=abs(cwtmatr).max(),
vmin=-abs(cwtmatr).max())
#plt.imshow(cwtmatr[:200,:], extent=[t[0], t[-1], widths[200], widths[0]], cmap='PRGn', aspect='auto', vmax=abs(cwtmatr).max(), vmin=-abs(cwtmatr).max())
plt.title('CWT')
#plt.xlabel('?')
#plt.ylabel('?')
plt.tight_layout()
plt.show()
def test_whiten(self):
# create noise with a glitch in it at 1000 Hz
noise = self.TEST_CLASS(numpy.random.normal(loc=1, size=16384 * 10),
sample_rate=16384,
epoch=-5).zpk([], [0], 1)
glitchtime = 0.5
glitch = signal.gausspulse(noise.times.value + glitchtime,
bw=100) * 1e-4
data = noise + glitch
# whiten and test that the max amplitude is recovered at the glitch
tmax = data.times[data.argmax()]
assert not numpy.isclose(tmax.value, -glitchtime)
whitened = data.whiten(2, 1)
assert noise.size == whitened.size
nptest.assert_almost_equal(whitened.mean().value, 0.0, decimal=4)
tmax = whitened.times[whitened.argmax()]
nptest.assert_almost_equal(tmax.value, -glitchtime)
def gen_test_data(plot=False):
""" Generate factor matrices which components will be easy to differentiate from one another
Parameters
----------
plot : bool
Returns
-------
fmat : list[np.ndarray]
core_values : np.ndarray
"""
t_A = np.linspace(0, 1, 500, endpoint=False).reshape(-1, 1)
t_B = np.linspace(0, 2, 10, endpoint=False).reshape(-1, 1)
t_C = np.linspace(-1, 1, 2 * 100, endpoint=False).reshape(-1, 1)
w_A = np.array([1, 2, 5]).reshape(-1, 1)
w_B = np.roll(w_A, 1)
w_C = np.array([0.3, 2, 0.7]).reshape(-1, 1)
A = np.sin (2 * np.pi * t_A * w_A.T)
B = signal.square (2 * np.pi * t_B * w_B.T)
C, _, _ = signal.gausspulse(t_C * w_C.T, fc=5, retquad=True, retenv=True)
fmat = [A, B, C]
core_values = np.array([1]*A.shape[1])
if plot:
for mode, factor in enumerate(fmat):
print("Mode-{} factor matrix shape = {}".format(mode, factor.shape))
fig, axis = plt.subplots(nrows=3,
ncols=1,
figsize=(8, 8)
)
axis[0].plot(t_A, A)
axis[0].set_title("Factor matrix A")
axis[1].plot(t_B, B)
axis[1].set_title("Factor matrix B")
axis[2].plot(t_C, C)
axis[2].set_title("Factor matrix C")
plt.tight_layout()
return fmat, core_values
def test_correlate(self):
# create noise and a glitch template at 1000 Hz
noise = self.TEST_CLASS(
numpy.random.normal(size=16384 * 64), sample_rate=16384, epoch=-32
).zpk([], [1], 1)
glitchtime = -16.5
glitch = self.TEST_CLASS(
signal.gausspulse(numpy.arange(-1, 1, 1./16384), bw=100),
sample_rate=16384, epoch=glitchtime-1)
# check that, without a signal present, we only see background
snr = noise.correlate(glitch, whiten=True)
tmax = snr.times[snr.argmax()]
assert snr.size == noise.size
assert not numpy.isclose(tmax.value, glitchtime)
nptest.assert_almost_equal(snr.mean().value, 0.0, decimal=1)
nptest.assert_almost_equal(snr.std().value, 1.0, decimal=1)
# inject and recover the glitch
data = noise.inject(glitch * 1e-4)
snr = data.correlate(glitch, whiten=True)
tmax = snr.times[snr.argmax()]
nptest.assert_almost_equal(tmax.value, glitchtime)
def test_save_loudest_tile_features():
# prepare input data
channel = GW.channels[0]
noise = TimeSeries(
numpy.random.normal(loc=1, scale=.5, size=16384 * 68),
sample_rate=16384, epoch=-34).zpk([], [0], 1)
glitch = TimeSeries(
signal.gausspulse(numpy.arange(-1, 1, 1./16384), bw=100),
sample_rate=16384, epoch=-1) * 1e-4
in_ = noise.inject(glitch)
_, _, _, qgram, _, _, _ = core.scan(
gps=0, channel=channel, xoft=in_, resample=4096, fftlength=8)
# test loudest tiles
channel.save_loudest_tile_features(qgram, correlate=glitch)
assert channel.Q == numpy.around(qgram.plane.q, 1)
assert channel.energy == numpy.around(qgram.peak['energy'], 1)
assert channel.snr == numpy.around(qgram.peak['snr'], 1)
assert channel.t == numpy.around(qgram.peak['time'], 3)
assert channel.f == numpy.around(qgram.peak['frequency'], 1)
assert channel.corr == numpy.around(glitch.max().value, 1)
assert channel.delay == 0.0
assert channel.stdev == glitch.std().value
from scipy import signal import matplotlib.pyplot as plt import numpy as np t = np.linspace(-1, 1, 200, endpoint=False) sig = np.cos(2 * np.pi * 7 * t) + signal.gausspulse(t - 0.4, fc=2) fig1,handler = plt.subplots() handler.grid(True) handler.plot(t,sig) # widths = np.arange(1, 31) cwtmatr = signal.cwt(sig, signal.ricker, widths) fig2,handler2 = plt.subplots() plt.imshow(cwtmatr, extent=[-1, 1, 1, 31], cmap='PRGn', aspect='auto',vmax=abs(cwtmatr).max(), vmin=-abs(cwtmatr).max()) #handler2.plt.show() plt.show()
from .. import qtransform
from ...table import EventTable
from ...segments import Segment
from ...timeseries import TimeSeries
__author__ = 'Alex Urban <*****@*****.**>'
# -- global variables ---------------------------------------------------------
# create noise and a glitch template at 1000 Hz
NOISE = TimeSeries(
numpy.random.normal(size=4096 * 10), sample_rate=4096, epoch=-5)
GLITCH = TimeSeries(
gausspulse(NOISE.times.value, fc=500)*10, sample_rate=4096)
DATA = NOISE + GLITCH
# global test objects
SEARCH = Segment(-0.25, 0.25)
QGRAM, FAR = qtransform.q_scan(DATA, search=SEARCH)
QSPECGRAM = QGRAM.interpolate()
# -- test utilities -----------------------------------------------------------
def test_far():
# test that FAR is better than 1 / Hubble time
assert FAR < 1 / (1.37e10 * 365 * 86400)
def gausspulse(self): return [ self.K * u for u in signal.gausspulse( t = self.t, fc = self.f) ]
# Plot real component, imaginary component, and envelope for a 5 Hz pulse, # sampled at 100 Hz for 2 seconds: from scipy import signal import matplotlib.pyplot as plt t = np.linspace(-1, 1, 2 * 100, endpoint=False) i, q, e = signal.gausspulse(t, fc=5, retquad=True, retenv=True) plt.plot(t, i, t, q, t, e, '--')
# -*- coding: utf-8 -*-
"""
Created on Wed Nov 25 18:13:27 2015
@author: NigmatullinR
"""
#пример связ с Гаусовским импульсом
import numpy as np
import pylab as plb
import scipy.signal as sgn
M=2 # decimation factor
t = np.linspace(-0.75, 0.75, 2 * 100, endpoint=False)#носитель на кот строится ГИ
g = sgn.gausspulse(t, fc=5, bw=0.3, retquad=False, retenv=False)#Гаус импульс
plb.plot(t, g)
plb.title('Original signal')
plb.figure()
plb.plot(abs(np.fft.fft(g)[:len(g)/2]))#преобр Ф
plb.title('Original signal FFT')
ind2=range(0,len(g),2)
g2=g[ind2]
plb.figure()
plb.plot(g2)
plb.title('Signal after decimation')
plb.figure()
def myfnc(t):
return sig.gausspulse(t, fc = 0.07)
from numpy.random import normal
from scipy.signal import gausspulse
from gwpy.timeseries import TimeSeries
# Generate a `TimeSeries` containing Gaussian noise sampled at 4096 Hz,
# centred on GPS time 0, with a sine-Gaussian pulse ('glitch') at
# 500 Hz:
noise = TimeSeries(normal(loc=1, size=4096*4), sample_rate=4096, epoch=-2)
glitch = TimeSeries(gausspulse(noise.times.value, fc=500) * 4, sample_rate=4096)
data = noise + glitch
# Compute and plot the Q-transform of these data:
q = data.q_transform()
plot = q.plot()
ax = plot.gca()
ax.set_xlim(-.2, .2)
ax.set_epoch(0)
plot.show()
sim_fixed.set_parameter("verbose", "0"); sim_spline.set_parameter("verbose", "0")
sim_fixed.set_print_debug(False); sim_spline.set_print_debug(False)
sim_fixed.set_parameter("sound_speed", "%f" % c0); sim_spline.set_parameter("sound_speed", "%f" % c0)
sim_fixed.set_parameter("phase_delay", "on"); sim_spline.set_parameter("phase_delay", "on")
sim_fixed.set_parameter("radial_decimation", "5"); sim_spline.set_parameter("radial_decimation", "5")
num_gpus = int(sim_fixed.get_parameter("num_cuda_devices"))
print "System has %d CUDA devices" % num_gpus
sim_fixed.set_parameter("gpu_device", "%d" % args.gpu_device_no)
sim_spline.set_parameter("gpu_device", "%d" % args.gpu_device_no)
print "Fixed simulator uses %s" % sim_fixed.get_parameter("cur_device_name")
print "Spline simulator uses %s" % sim_spline.get_parameter("cur_device_name")
# define excitation signal
t_vector = np.arange(-16/args.fc, 16/args.fc, 1.0/args.fs)
samples = np.array(gausspulse(t_vector, bw=args.bw, fc=args.fc), dtype="float32")
center_index = int(len(t_vector)/2)
demod_freq = args.fc
sim_fixed.set_excitation(samples, center_index, args.fs, demod_freq)
sim_spline.set_excitation(samples, center_index, args.fs, demod_freq)
# create big scan sequence with all M-mode beams (for the spline algorithm)
origins = np.empty((args.num_beams_total, 3), dtype="float32")
directions = np.empty((args.num_beams_total, 3), dtype="float32")
lateral_dirs = np.empty((args.num_beams_total, 3), dtype="float32")
y_axis = np.array([0.0, 1.0, 0.0])
lateral_dir = np.cross(y_axis, direction)
for beam_no in range(args.num_beams_total):
origins[beam_no, :] = origin
directions[beam_no, :] = direction
lateral_dirs[beam_no, :] = lateral_dir
fc_upp=1000
a, b = truncparms(fc_low, fc_upp, fc_mu, fc_sigma)
fcs = stats.truncnorm.rvs(a, b, loc=fc_mu, scale=fc_sigma, size=Nwaveforms)
# bandwidth
bw_mu=0.005
bw_sigma=0.001
bw_low=0.0001
bw_upp=0.1
a, b = truncparms(bw_low, bw_upp, bw_mu, bw_sigma)
bws = stats.truncnorm.rvs(a, b, loc=bw_mu, scale=bw_sigma, size=Nwaveforms)
# populate catalogue
win = lal.CreateTukeyREAL8Window(len(time_axis), 0.1)
for i in xrange(Nwaveforms):
catalogue[:,i] = win.data.data*signal.gausspulse(time_axis, fcs[i], bws[i])
catalogue[:,i] /= np.sqrt(np.dot(catalogue[:,i], catalogue[:,i]))
#
# Chirps
#
chirp_times = np.arange(0, datalen, 1.0/Fs)
idx = chirp_times<=0.5
win = lal.CreateTukeyREAL8Window(int(sum(idx)), 0.1)
tau=0.5
f0=100
t1=0.5
f1_mu = 500
f1_sigma = 100
Just half sampling of group3 otus.
'''
ts_g6_otus = ts_g3_otus.take(arange(0,74,2),1).astype(int)
###############
# group7 otus #
###############
'''
These otus will include a pulse for some of the samples. The pulses are
all shifted by one sample, and also includes their envelope. Here, the
center frequency is 1Hz
'''
#make pulse of 50 points, with amplitude 200
t = linspace(-3, 3, 1*50, endpoint=False)
real, envelope = gausspulse(t, fc=1, retenv=True)
real = 200*(real+1)
envelope = 200*(envelope+1)
# plt.plot(t, i, t, e, '--')
# plt.show()
samples = linspace(0, 199, 200)
otu_real =[[0 for a in range(200)] for k in range(200)]
otu_envelope =[[0 for a in range(200)] for k in range(200)]
#insert pulse into 150 point vector of average amplitude
#at different indices for total length of 200
for j in samples:
z = [200]*150
j = int(j)
for x in real:
import numpy import scipy import matplotlib.pyplot as plt x=numpy.linspace(-1, 1, 1000) t=numpy.linspace(-3*numpy.pi,3*numpy.pi,1000) from scipy.signal import chirp, sawtooth, square, gausspulse plt.figure() plt.subplot(221) plt.plot(x,chirp(x,f0=100,t1=0.5,f1=200)) plt.ylim([-2,2]) plt.subplot(222) plt.plot(x,gausspulse(x,fc=10, bw=0.5)) plt.ylim([-2,2]) plt.subplot(223) plt.plot(t,sawtooth(t)) plt.ylim([-2,2]) plt.subplot(224) plt.plot(t,square(t)) plt.ylim([-2,2]) plt.show()
def do_simulation(args):
if args.use_gpu:
sim = RfSimulator("gpu")
sim.set_parameter("gpu_device", "%d"%args.device_no)
gpu_name = sim.get_parameter("cur_device_name")
print "Using device %d: %s" % (args.device_no, gpu_name)
else:
sim = RfSimulator("cpu")
sim.set_parameter("verbose", "0")
with h5py.File(args.h5_file, "r") as f:
scatterers_data = f["data"].value
sim.add_fixed_scatterers(scatterers_data)
print "The number of scatterers is %d" % scatterers_data.shape[0]
# configure simulation parameters
sim.set_parameter("sound_speed", "1540.0")
sim.set_parameter("radial_decimation", "10")
sim.set_parameter("phase_delay", "on")
sim.set_parameter("noise_amplitude", "%f" % args.noise_ampl)
# configure the RF excitation
fs = 80e6
ts = 1.0/fs
fc = 5.0e6
tc = 1.0/fc
t_vector = np.arange(-16*tc, 16*tc, ts)
bw = 0.3
samples = np.array(gausspulse(t_vector, bw=bw, fc=fc), dtype="float32")
center_index = int(len(t_vector)/2)
sim.set_excitation(samples, center_index, fs, fc)
# define the scan sequence
origins = np.zeros((args.num_lines, 3), dtype="float32")
origins[:,0] = np.linspace(args.x0, args.x1, args.num_lines)
x_axis = np.array([1.0, 0.0, 0.0])
z_axis = np.array([0.0, 0.0, 1.0])
directions = np.array(np.tile(z_axis, (args.num_lines, 1)), dtype="float32")
length = 0.06
lateral_dirs = np.array(np.tile(x_axis, (args.num_lines, 1)), dtype="float32")
timestamps = np.zeros((args.num_lines,), dtype="float32")
sim.set_scan_sequence(origins, directions, length, lateral_dirs, timestamps)
# configure the beam profile
sim.set_analytical_beam_profile(1e-3, 1e-3)
frame_sim_times = []
for frame_no in range(args.num_frames):
start_time = time()
iq_lines = sim.simulate_lines()
frame_sim_times.append(time()-start_time)
if args.save_simdata_file != "":
with h5py.File(args.save_simdata_file, "w") as f:
f["sim_data_real"] = np.array(np.real(iq_lines), dtype="float32")
f["sim_data_imag"] = np.array(np.imag(iq_lines), dtype="float32")
print "Simulation output written to %s" % args.save_simdata_file
print "Simulation time: %f +- %f s (N=%d)" % (np.mean(frame_sim_times), np.std(frame_sim_times), args.num_frames)
if args.pdf_file != "" and not args.visualize:
import matplotlib as mpl
mpl.use("Agg")
if args.pdf_file != "" or args.visualize:
import matplotlib.pyplot as plt
num_samples, num_lines = iq_lines.shape
plt.figure(1, figsize=(18,9))
plt.subplot(1,2,1)
plt.plot(np.real(iq_lines[:, num_lines/2]))
plt.title("Middle RF line")
plt.subplot(1,2,2)
plt.imshow(np.real(abs(iq_lines)), aspect="auto", interpolation="nearest")
if args.pdf_file != "":
plt.savefig(args.pdf_file)
print "Image written to disk."
if args.visualize:
plt.show()